Same week, three labs, one quiet pattern: Anthropic, OpenAI, and DeepSeek all made it harder or more expensive to use their models, and none of them described it that way. Plus a five-year-old automation startup shut down because the platforms it depended on absorbed its entire reason for existing, and Google undercut OpenAI on live-audio pricing partly just by publishing a price at all.

17%
Net cut to Claude Code's weekly usage allowance versus the summer boost, once the math is done
2x
DeepSeek's new peak-hour price compared to off-peak
$1.38
Roughly what an hour of Gemini 3.8 Live audio costs, against $3 or more for OpenAI's competing model

A Permanent Increase That's Actually a Cut

Anthropic ended a temporary 50% weekly usage boost for Claude Code that had been running since May 13, replacing it with a permanent 25% increase over the original baseline. Run the numbers and that's a net reduction of roughly 17% against the allowance Pro, Max, Team, and seat-based Enterprise customers had gotten used to over the summer. Session windows are unchanged, five hours for Max 5x and 20x plans.

Anthropic's framing was that it was making a temporary boost permanent. That's technically accurate. It's also a cut, and Claude Code's user base is exactly the crowd most likely to notice a weekly ceiling: people running it as a daily agentic coding tool, not casual users who'd never bump into a limit either way.

Three Labs, One Week, One Pattern

Anthropic's move didn't happen in isolation. In the same window, ChatGPT Pro closed to new signups, and DeepSeek began pricing peak hours at double its off-peak rate.

Same Week, Different SpinAnthropic: a "permanent increase" that nets out to less than the summer allowance. OpenAI: a paid tier simply closed to new customers, no cut announced because there's nothing left to cut, access itself is the limit. DeepSeek: the same compute, priced differently depending on when you show up.

Three different mechanisms, the same underlying problem: demand for frontier model access is outrunning the compute available to serve it, and price and access are the two levers labs have to ration that gap without using the word rationing. None of the three announcements framed it as a capacity problem. All three functionally are one.

A Zapier Alternative Didn't Survive the Platforms It Competed With

Relay, a five-year-old AI-powered workflow automation tool built as an alternative to Zapier, shut down entirely on September 15. The product let users automate email and task workflows using AI agents, a genuinely useful niche a few years ago. Its own explanation for closing was straightforward: once OpenAI, Google, and other larger platforms built equivalent automation directly into their own tools, a five-year-old standalone product no longer had a clear reason to exist on its own.

It's a small story next to the usage-rationing pattern above, but it's the same underlying dynamic from the other direction: as the big platforms absorb more of the workflow layer directly, smaller tools built on top of them lose their footing one at a time.

Gemini Undercuts GPT-Live-1, Partly Just by Naming a Price

Google's Gemini 3.8 Live launched at roughly $1.38 an hour for audio, $0.005 per minute of input and $0.018 per minute of output. OpenAI's competing GPT-Live-1 shipped to the API on September 10 at $3 or more an hour, and without OpenAI ever publishing a rate card of its own.

Part of what makes Gemini's move effective isn't just the lower number. It's that Google published one at all. A model with a clear, public price is an easier default for a developer comparing options than a competitor's model without one, regardless of what that unpublished number would eventually turn out to be.

None of these four items is individually a major story. Together, they're a useful snapshot of where the industry actually is right now: labs managing real capacity constraints through pricing and access rather than headline capability announcements, and smaller tools discovering in real time how much of their market the platforms they built on top of can simply absorb.

Sources: AI Pulse · Spilling the Tea · workplaceai.ai. Claude Code usage changes and the three-lab pricing pattern: AI News Today digest, September 16, 2026. Relay's shutdown: TechCrunch, "The AI graveyard: a running list of projects and startups that didn't make it," September 15, 2026. Gemini 3.8 Live and GPT-Live-1 pricing: AI News Today digest, September 16, 2026. Every figure above is attributed to its original reporting; none is a WorkplaceAI study.